Contrarian Minds: Jim Mitchell

Leading the charge to create a peta-scale computer.

By Al Riske

13.May.04--In Cambridge, Ontario -- a farming community about 60 miles from Toronto -- Jim Mitchell grew up wanting to do something that would make a difference in the world.

"Maybe it was the unbounded optimism of the young. I don't know why," he says now. "I just fantasized about doing something that would make the world better."

He dreamed up the first WATFOR compiler when he was still an undergraduate at the University of Waterloo, designed the Mesa programming language at Xerox PARC in 1973, and led research into the first distributed, object-oriented operating system, Spring, when he joined Sun in 1988.

A compact, energetic man who built model airplanes as a boy and now flies real ones, Mitchell is a contrarian thinker and a risk taker. Recognition and financial rewards remain secondary to the intellectual stimulation he refers to as "geek candy."

At 60, Mitchell has handed over the reins of SunLabs and is taking on a new challenge -- directing Sun's efforts to design the technologies for a peta-scale supercomputer capable of executing hundreds of thousands of software threads simultaneously.

"It's as if my career has prepared me for this project," Mitchell says. "The hardware, software, and people-ware challenges in developing such a system make it extremely exciting and fun.

"It struck me, too, that I have worked in most of these areas -- physics, networking, operating systems, language design, virtual machines, user interfaces, file systems. So it feels like an awfully good fit," he says.

Thanks to a three-year, $50-million contract with DARPA (the U.S. government's Defense Advanced Research Projects Agency), Mitchell is leading a dream team of innovators that comprises some of Sun's most accomplished researchers and engineers, including both senior people such as Sun Fellow Dr. Ivan Sutherland and younger ones such as Dr. Robert Drost.

Many of those people, like Mitchell himself, are "outside the box" thinkers. In fact, one of the key ideas behind the project is a novel approach to computer system construction -- having chips interact with each other by simply placing them in close enough proximity to communicate wirelessly. This makes for more dense packaging, shorter distances between processor chips and memory chips, and a very high rate of communications to break today's memory-processor bottleneck.

"I came to Sun because it was a company that embraced innovation at all levels," Mitchell says. "Working with these creative people on new, promising ideas is what keeps me here."

Sun's reputation for contrarian notions is well known, but how did we get there?

"We got that reputation because we have the right culture," Mitchell says. "So the real question is 'How did we get that culture?'

"There's a short answer: Scott McNealy and the company's founders, people like Bill Joy and Andy Bechtolsheim.

"When I joined the company in 1988, we were the young upstarts (although that was the year Sun first made over $1 billion in revenue). We had hired the best and brightest from Berkeley, Stanford, PARC, and so on. And the people we hired wanted to do things that would change the world. So we kind of collected a set of people who defined the initial culture, and we had Scott, hockey-player Scott, highly competitive, kick-butt-and-have-fun Scott, always looking to say, 'The fun here is to find these Goliaths and overtake them.'

"We passed HP twice. Once as HP; then they bought Apollo, which was actually a bigger competitor. But they had difficulties merging the Apollo and HP products and cultures, and we passed them again in terms of market share -- in those days, engineering workstations. If you look at HP today after adding Compaq, they are bigger still, which just gives us a shot at overtaking them yet again.

"So there's that competitive thing, Scott's competitiveness. 'Go gang, go gang. Beat those guys. The enemy is them. We're at war.' All that stuff, but coupled with the smarts and the will to dare to do things that change the industry, things like Java, like thin-client desktops, like chip multi-threading."

Into this culture Sun attracted a set of the best and the brightest, not necessarily those easiest to manage. "Engineers who are easy to manage tend to be kind of average," Mitchell says. "A small number of creative, motivated engineers can trump a large army of engineers any day -- it's been proven again and again. Sun is full of such small, focused, creative teams."

His own most famous contrarian project may be the Java Community Process. "I got a few scars," he says.

The result, though, has been nothing less than the best standard-setting process in the world. "It's done over 200 effective standards in the same time most 'de jure' standards bodies have done a few dozen less effective standards each," Mitchell says.

"Out of the JCP you not only get the standard -- the document that specifies what it does -- but you get a reference implementation [actual software] and a compatibility test. Typically a company will take the reference implementation, change the parts where their company adds value -- say, better access to their database or something. Then they'll use the test program to make sure they got it right, and when two such companies produce their first products based on that Java specification, they tend actually to be compatible."

With other standards bodies, that doesn't happen, Mitchell says. All you get is a description of the standard. No code. No test suite. And there's often no requirement that you implement all the features, so you end up with incompatible products.

"There are many examples," Mitchell says. "Back in the 1970s there was an attempt at high-speed data communications called x25, and the first x25 systems (think of them as modems) from even one company weren't compatible. You couldn't buy two different models and have them work together. It took three generations of products before you could interoperate between companies."

Will his new project be as successful? Can his team really find a way to build a supercomputer without wires?

It's a monumental task, and not without risk. Designing the hardware is a huge challenge. Creating the software that makes it easier for programmers to write applications that can release the computational power of the system is also a big challenge. But for Mitchell, "doing all this research in just a few years only increases the sense of urgency and excitement."

Why not lash a bunch of smaller computers together into a grid? "Because of the speed of light," Mitchell says.

Huh?

"When signals travel between computers in a grid, you're going long distances over a local area network or farther," Mitchell explains, "and that, relative to what we're talking about, is painfully slow, even if you've got gigabit Ethernet. All the data for the program has to go over the network, because there's no shared memory in a grid system.

"Sun's HPCS system is being designed to handle very big problems needing lots of shared data -- things like weather simulation, computational biology, finite element analysis, physical simulations," he says.

"We are developing technologies in this program that are valuable to Sun's products outside the realm of supercomputers," Mitchell says. "It's obvious with proximity communication, but I think even with the management software, novel execution models, and so on -- that's got value for our current and future products."

Growing up in Ontario, building balsa airplanes he designed himself -- and driving his mother crazy with all the pin holes he poked in the wood floor of his bedroom to lay the wings out flat -- Mitchell could never have imagined where life would take him. But here he is, doing work that matters -- to Sun, the U.S. government, and himself.